MxGPS: Multiplex Graph Transformers for a Power Grid Foundation Model

The paper introduces MxGPS, a multiplex graph transformer with a shared node encoder and K task-specialized GPS branches. It jointly trains on Static State Estimation (SSE) and AC Power Flow (PF) using self-supervised pre-training and multi-task fine-tuning. In 3-fold sliding-window cross-validation across four unseen topologies with 14, 24, 162, and 300 buses, MxGPS achieved a 0% boundary violation rate for zero-shot PF. Its performance degradation under topology shift was 39%, compared with 190% to 1400% for models with lower in-distribution PF error. The model uses 1.6M parameters, 12x fewer than the cited GridFM baseline.